As Director of the Scientific Data Division at Lawrence Berkeley National Laboratory, Ana Kupresanin leads scientists and engineers who develop the methods, software, workflows, and infrastructure needed to make scientific data usable, reliable, and reusable for science and AI. The division works across the scientific data lifecycle, helping researchers organize, curate, manage, access, analyze, and reuse data, while also developing machine learning methods, high-performance computing workflows, and partnerships with domain scientists across disciplines.
Kupresanin is a statistician and a Fellow of the American Statistical Association. Before joining Berkeley Lab in 2023, she spent more than a decade at Lawrence Livermore National Laboratory, where she held scientific and leadership roles and worked with researchers across fields to develop statistical methods, analyze complex data, and address uncertainty quantification problems.
That background shapes how she thinks about AI for science. Scientific data are not generic inputs to a model. They come from experiments, simulations, instruments, and observations, each with its own assumptions, limitations, uncertainties, and context. Kupresanin's work focuses on bringing statistical thinking, machine learning, and data infrastructure together so that AI systems can be more reliable, interpretable, and useful for scientific discovery.
In this Q&A, Kupresanin discusses where AI for science is headed and why Berkeley Lab's strengths in data, computing, facilities, and scientific expertise are central to the Department of Energy's Genesis Mission, a national initiative to advance AI for scientific discovery and address challenges in science, energy, and national security.
Q: From your vantage point, what is the "big picture" on AI and its potential for science and delivering advances?
Kupresanin: AI has the potential to change not only how fast we do science, but how we formulate and test scientific ideas. It can help researchers work with data and complexity at a scale that would otherwise be very difficult, connecting experiments, simulations, theory, and observations in new ways. But, for science, AI is not about using the largest or newest model just because it is fashionable. The real opportunity is to build AI systems that are grounded in scientific data, domain knowledge, uncertainty, and physical constraints. If we do that well, AI can help accelerate discovery while making results more reproducible, interpretable, and trustworthy.
Q: What are the most significant ways Berkeley Lab is contributing to the Genesis Mission
Kupresanin: Berkeley Lab's contribution to Genesis is the scientific ecosystem it brings together. The Lab has world-class expertise in scientific data, high-performance computing, high-speed networking, user facilities, automated laboratories, simulations, AI models, software, and domain science. That combination is essential because AI for science depends on connecting models with the data, infrastructure, facilities, and expertise that make them useful and trustworthy. Berkeley Lab can help make scientific data usable for AI, connect that data to advanced computing and experimental facilities, and support new workflows in which AI, experiments, simulations, and researchers work together.
Q: How does Berkeley Lab's expertise and capabilities make the lab well positioned to contribute to the Genesis Mission?
Kupresanin: Berkeley Lab is well positioned for Genesis because it combines many of the resources needed for AI-enabled science in one place: facilities that generate unique data, advanced computing and networking capabilities, expertise in simulation and software, people who work on data management, statistics, uncertainty, and AI, and researchers who are experts in their fields. The power of Genesis is in synthesizing these pieces in a connected scientific workflow, from experiments and simulations to data preparation, model development, analysis, interpretation, and reuse.
From my perspective as a statistician and director of the Scientific Data Division, the data piece is especially important. Scientific data are not generic inputs to AI. They carry context: how they were generated, what assumptions went into them, what uncertainties they have, and what they can reliably be used for. Berkeley Lab can help make that context usable for trustworthy AI for science.
